|
Activity Number:
|
112
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 9, 2004 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #301671 |
|
Title:
|
A Joint Longitudinal and Survival Model that Flexibly Incorporates Nonproportional Hazards Treatment Effects
|
|
Author(s):
|
Daniel L. Gillen*+ and Sebastien Haneuse
|
|
Companies:
|
University of Chicago and University of Washington
|
|
Address:
|
5841 South Maryland Ave., Chicago, IL, 60637,
|
|
Keywords:
|
survival ; nonproportional hazards ; longitudinal ; Bayesian
|
|
Abstract:
|
Time-to-event is a common outcome in many prospective studies. In addition, researchers may also monitor additional biologic endpoints that are hypothesized to be associated with survival as patients are continuing to be followed. Recently, Brown and Ibrahim (2003) considered a Bayesian semiparametric hierarchical model for simultaneously modeling survival and longitudinal data in a proportional hazards setting. We consider an extension of the Brown and Ibrahim model which flexibly accounts for nonproportional hazards covariate effects on survival using a conditional hazards survival model similar to that introduced by McKeague and Tighiouart (2000). In the conditional hazards survival model, the baseline log-hazard and covariate effects are estimated via step functions, forming a first-order autoregressive process, while the grid of jump times that define these step functions form a time-homogenous Poisson process. The model is fit using a Metropolis-Hastings-Green algorithm. We demonstrate the method using a clinical example in which time to mortality is modeled as a function of longitudinally measured serum albumin in end-stage renal disease patients.
|
